Ian Klosowicz
The statistics that show up in analyst interviews are 8 ideas: mean versus median, spread, percentiles and outliers, correlation versus causation, sampling and bias, hypothesis tests with p-values, A/B tests, and confidence intervals. If you can explain each one in plain language and apply it to a business question, you're covered.
Interviewers want to know whether you'll report a misleading number to a stakeholder, and whether you can say why a result might be wrong. That's a judgment test, and it's learnable in a couple of weeks.
I don't have a degree in this. I learned what I needed while working somewhere else and got hired anyway. Here's the short version of what actually gets asked, and what only lives in textbooks.
Eight ideas cover almost every statistics question in an entry-level analyst interview, and this is the order to learn them in:
The first 5 are the most common. The last 3 depend more on the company, and they matter most for roles near product or marketing. If you're deciding how deep to go overall, when statistics knowledge matters and when it doesn't covers that side of it.
Use the median instead of the mean when a few extreme values pull the average away from what's typical. Interviewers love this one because it's the simplest way to see whether you think before you report a number.
Here's a quick example. Say 5 orders came in at $20, $22, $25, $28, and $400. The mean is $99, and the median is $25. If you tell a stakeholder the average order is $99, you've described an order almost nobody placed.
The 3 measures work like this:
A strong interview answer names the tradeoff out loud: "I'd check the distribution first, and if it's skewed I'd report the median alongside the mean." That one sentence signals more judgment than any formula.
Report spread alongside the average, because 2 datasets can share a mean and behave nothing alike. Standard deviation is the usual measure, and a small one means values cluster near the average while a large one means they scatter.
For a roughly bell-shaped distribution, about 68% of values fall within 1 standard deviation of the mean and about 95% within 2. You don't need to compute it by hand. You need to explain what it tells a stakeholder, like how consistent delivery times are.
Percentiles are often more useful in practice. The 90th percentile of response time tells you what the slowest 10% of customers experience, and that's usually the number someone actually cares about.
Outliers get their own question in most interviews. The right approach has 3 steps:
Never silently delete an outlier. Interviewers want to hear that you ask what caused it before you decide what to do with it.
Correlation means 2 things move together, and it doesn't prove that one causes the other. This is one of the most common statistics questions in analyst interviews, so have a clear answer ready.
A clean way to explain it is an example. Ice cream sales and sunburns rise together in summer. Neither causes the other. Hot weather drives both. In business terms, customers who open your emails might buy more, but that doesn't mean the emails caused the purchases, because engaged customers do both.
When you hit this question, a good answer covers 3 points:
That last point is what separates a good answer from a memorized one.
A sample is only useful if it represents the group you want to describe. If it doesn't, your result is biased no matter how large the sample is, and interviewers will probe whether you notice.
Watch for these common problems:
A good habit is to ask "who isn't in this data?" before you trust any conclusion. It's a short question, and it shows you think about where the numbers came from instead of just what they say.
A p-value is the probability of seeing a result at least this extreme if nothing is actually going on. A small one, conventionally below 0.05, means the result would be unusual under pure chance. It does not tell you the chance your hypothesis is true.
That distinction is the most common place candidates slip. The plain-language version goes like this:
You should also know the 2 ways a test can go wrong. A false positive means you declared a difference that isn't real. A false negative means you missed one that is. Most entry-level questions stop there. Skip the formulas for specific test types until a job requires them.
An A/B test compares 2 versions by randomly assigning users to each, and the key interview point is that a statistically significant result isn't always one that matters to the business.
With enough users, a tiny difference becomes statistically significant. A lift of 0.1 percentage points in conversion can clear the bar with millions of visitors and still not justify the engineering cost. Statistical significance asks whether the difference is real. Practical significance asks whether it's worth acting on.
A solid answer to "how would you run an A/B test" covers these steps:
That last point matters. Checking results repeatedly and stopping at the first win inflates false positives, and interviewers who run experiments will notice if you skip it.
A confidence interval is a range that expresses how uncertain an estimate is, and the useful skill is explaining one without jargon. If a survey says 40% of customers prefer a feature with a margin of error of 3 points, the honest statement is "somewhere between 37% and 43%."
Two things to remember:
You'll get credit for the stakeholder-friendly version: report a range, not a single number, so people don't treat an estimate as a fact.
Skip the heavy theory, because entry-level interviews almost never reach it. These topics eat weeks and rarely come up:
If a job posting asks for modeling or experimentation specifically, treat that as a signal to go deeper. Otherwise, the 8 ideas above are the target.
Answer in 3 beats: the plain-language definition, a business example, and the pitfall. That structure keeps you from rambling, which is what usually sinks a good candidate.
I failed roughly my first 10 interviews. I rambled, I blacked out partway through answers, and I didn't know how to structure a response when I had no prior experience to point to. A repeatable structure fixes most of that. Practice each of the 8 ideas out loud with a timer until you can answer in under a minute.
If you want to see what the interview loop looks like after a rough one, what to do after failing a data analyst technical interview walks through the recovery.
Month 3 of the program at Analyst Hive covers technical and behavioral interview prep.
How much statistics do I need for a data analyst interview?
Enough to explain 8 ideas in plain language: mean versus median, spread, percentiles and outliers, correlation versus causation, sampling bias, p-values, A/B tests, and confidence intervals. You don't need to derive formulas. Interviewers want to see judgment about when a number might mislead, and that comes from understanding the concepts well enough to apply them.
Do analyst interviews ask about p-values?
Often, especially at companies that run experiments. The usual question asks what a p-value means. The safe answer: it's the probability of seeing a result this extreme if there were no real effect, and a small one suggests the result isn't just chance. Avoid saying it's the probability your hypothesis is true.
What is the most common statistics question in an analyst interview?
Correlation versus causation is one of the most common, along with mean versus median. Both test whether you think critically about a number before reporting it. Prepare a concrete business example for each, and practice saying them out loud so the answer sounds natural instead of recited.
Do I need to know how to calculate statistics by hand?
Rarely. Most interviews care about interpretation, not arithmetic. You should be able to calculate a mean and a median, and describe what standard deviation measures. Beyond that, expect to explain results and decisions rather than work through formulas on a whiteboard.
Should I learn statistics before SQL?
No. SQL appears in more job postings and gets tested more often, so it comes first. Statistics concepts are quick to pick up alongside your SQL and BI practice, and most candidates only need the 8 ideas in this post. Learn them in the weeks before interviews start.
How do I practice statistics for interviews?
Take a real dataset and compute the mean, median, and spread for one column. Look at the outliers and decide what you'd do. Then explain each result out loud as if talking to a stakeholder. Speaking the answer is the part most people skip, and it's the part that matters on the day.
If you want the full day-by-day plan, including interview prep, join Analyst Hive and work through it in order.
